Replies posted by jcytam
Hi Stefano The slower Clock of the Intel Xeon Processor 2.40GHz 8-core Haswell E5-2630 v3 does have a overall impact to our benchmark and is not the optimal choice to run PG Workload. However, our cluster is hosting a mixed-workload environment, we favoured the numbers of core rather than the clock. Cheers Jacky
Hi Kirkkak thanks for sharing your experience. We are a small company therefore, we are using our 6k Series Cluster for mix work load of server virtualizations. Our PostgreSQL is service Data for a Data Warehouse like Web Application. Together with Nutanix's Support Team we reached relative good performance outcome. I would like to share our setup here: our 3 Nodes 6k Cluster Setup, each Node has: 2 x Intel Xeon Processor 2.40GHz 8-core Haswell E5-2630 v3 20M Cache 12 x 32GB DDR4 Memory Module 5x 4TB 3.5" HDD 1 x 1.2TB 3.5" SSD 1 x 10GbE Dual SFP+ Network Adapter Virtual Machine: Ubuntu 16.04 64 Bit machine 8vCPU, 24GB RAM (allthought Nutnaix recommended 2-3xDB Size, our DB Size is 80GB, which leads to a optimal size of 160+ GB RAM. Since this cluster is not just dedicated for PG Workload we can not affort this huge amount of RAM) OS on separate controller (paravirt) OS Disk 0:0 PG data is on separate controller (paravirt) with multiple Disks 1:0-3 data disk -2:0-3 wal disk Lin
The Other Setup we compared: Test System 1 Desktop Computer: 1*Intel Core i5-4570-T CPU @ 3.20GHz, 12GB RAM, Intel SSD 256 GB. Test System 2 current production virtualized PostgreSQL Server running on ESX Host with NetApp NFS Storage: ESX Host IBM X3650 M4 with 2*Intel Xeon E5-2690v2 @ 3.00 GHz, 2*10GiE NetApp FAS2240-2 with 22*550GB SATA Disks + 4*200GB SSD Disks as Flash Pool, 2*10GiE PostgrESQL Server: 4vCPU, 16 GB RAM, PostgreSQL DB mounted directly from the NetApp Filer via NFS with mount options rw,hard,ac,nointr,rsize=32768,wsize=32768,bg,nfsvers=3,tcp,sec=sys.
Just want to share our experience so far with the community: Nutanix's Suggestions indicated as * and our Setup as >: * DB, Logs and wal on different vDisks>Currently, we just separate the OS+Postgresql Bineries and DB into separate vDisks.OS+PgSQL on a Thin Provisioned vDisk on a Deduped ContainerPostgreSQL DB on 4vDisks on a line-compressed Container, the 4 vDisks are aggregated into a single LVM Volume and consumed by PostgreSQL.* RAM 2-3X DB Size>We are a small company, we just cannot affort to allocate that much of Memory dedicated just a Database Server and our workload is not that high. Totally we have 1.13 TB of RAM in our Nutanix Cluster, the current Database Size is 80 GB. We allocated 24 GB RAM to this PostgreSQL Server.* The wal_buffers can be bumped up to 16MB>Our wal_buffer is already set to 16MB* If possible, disable synchronous_commit - Disabling this will disable the guarantee that the data was committed to disks, but will not corrupt the db.>How could we ensure th
Login to the community
Login with your account
Enter your username or e-mail address. We'll send you an e-mail with instructions to reset your password.